Azure AI Services at Scale for Cloud, Mobile, and Edge

Chapter 3. Train, Tune, and Deploy Models with Azure Machine Learning, ONNX, and PyTorch

In the previous chapter, we tried to cover the full range of AI tools and services available from Microsoft. Now let’s focus on how you can use the Azure Machine Learning cloud service to build and train your own models, using familiar machine learning frameworks and a mix of Azure and Visual Studio tooling. We’ll be looking at how you can use the popular PyTorch machine learning framework as well as how you can export trained models as ONNX for use with local inferencing runtimes, like ML.Net.

Understanding Azure Machine Learning

Microsoft’s approach to machine learning is to target different products to different groups of users, with different skill levels and different expectations for the technologies they’re using. At one end of the scale is the Power Platform’s AI Builder’s task-focused low-code connectors (we’ll look at those in Chapter 6), while at the other is Azure Machine Learning. Designed for experienced data scientists, it provides a cloud-based development environment where you can design, train, run, and manage machine learning models using popular frameworks.

The Azure Machine Learning environment is best thought of as a set of tools that all address the same backend model hosting infrastructure but that can be mixed and matched to fit with the way you want to work. If you’re new to advanced machine learning development, you can work with a drag-and-drop designer to build ...